--- title: ZADO-01 — AI Knowledge Curator slug: zado role: AI Knowledge Curator function: Operations seniority: Mid verticals: - "SaaS" - "Engineering" tier: professional monthlyPrice: $500/month publishedAt: "2026-05-04T00:00:00Z" canonical: "https://fidelic.ai/agents/zado" --- # ZADO-01 — AI Knowledge Curator *Cited Slack answers* Handles internal Slack Q&A with citations from your approved docs, and drafts the missing doc when the same question lands three times. ## Why it matters Senior people are the documentation. ZADO’s job is to relieve them — not by being smarter, but by surfacing what they already wrote, with citations. ## Capabilities - Watches approved Slack channels and Drive folders for repeated questions - Answers in-thread with cited sources from approved docs - Flags patterns where a written doc would replace recurring back-and-forth - Drafts the missing doc when the same question lands three times - Maintains the index of canonical answers as the team’s vocabulary shifts ## How it works ZADO listens for question patterns in approved Slack channels: who-asked-what, frequency, the people pinged for the answer. When a pattern emerges, ZADO answers in-thread with citations to the source doc — if a source exists. If not, ZADO drafts the missing doc and routes it to the owner. Citations are not paraphrase. ZADO never cites a source it hasn’t indexed. If the answer requires synthesis across multiple docs, ZADO names them all and flags it for the human owner rather than writing a new claim. ## The edge ZADO refuses to bullshit. If the answer isn’t in the approved docs, ZADO says so explicitly, lists the closest sources, and routes to the owner. Most knowledge bots fabricate confidence; ZADO publishes the missing-doc list. ## The hook The byproduct of running ZADO is a constantly-growing index of what your team actually doesn’t have written down. After three months you have a documentation backlog — ranked by frequency, ready to triage. ## Sample output ### Sample cited Slack answer (in-thread, anonymized) Question (from #engineering, B. Patel): “When is it OK to bypass the deploy gate for a hotfix?” ZADO answer: “Bypassing the deploy gate is allowed for P0 incidents (customer-data exposure, total outage, security CVE) per the Deploy Gate Policy [link to Drive doc, last updated 2026-03-12 by S. Mehta]. The bypass requires a on-call approver and a postmortem within 48 hours per the Postmortem SOP [link]. For P1 and below, follow the standard gate. If you’re unsure, this question landed in Slack 4 times in the last 60 days — the owner of the Deploy Gate Policy is S. Mehta.” ZADO never paraphrases without citation. If the policy doc didn’t cover P1, ZADO would say: “The policy doesn’t address P1; routing this to S. Mehta to confirm and update the doc.” All names and details in this sample are fabricated for illustration. ## Evidence - Pre-launch — public beta planned for Q2 2026. - Built on the Fidelic AI EvalOps test suite — citation-fidelity tests gate every release. - Inherits new base models and new agent skills automatically — pushed to every Fidelic agent the moment they ship. Same shape as a SaaS update; no upgrade purchase, no version pinning. ## Safeguards ### What ZADO-01 will not do Answer in channels or from docs you haven’t authorized at intake. ZADO’s scope is exactly what you grant access to. Make policy, strategy, or organizational decisions. ZADO surfaces what already exists and flags gaps; humans own the calls. Speak to customers. ZADO is an internal-only role. Share across team boundaries. If your engineering team’s docs are off-limits to sales, ZADO honors that boundary. ## Worst-case behavior Refuses to answer when sources are missing rather than guessing. Failure mode is silence on questions ZADO can’t cite — those route to the the workflow owner. ## Day 1 / Week 1 / Month 1 - **Day 1:** Indexes approved Slack channel history (last 60 days) and Drive folders. First clarifying questions on team-boundary rules and source-doc owners land in DMs. - **Week 1:** First in-thread cited answer ships. The team reviews and confirms ZADO’s tone; the citation format is locked. - **Month 1:** Recurring-question patterns are stable. The first missing-doc drafts land in the relevant owner’s inbox. The 30-day success metric has its first reading. ## 90-day success criterion By day 30, the senior team is asked the same recurring question half as often as before — measured against the channel’s baseline ping volume. ## Resume **Background.** ZADO-01 is a Fidelic AI template configured for internal Slack Q&A with strict citation-fidelity. Claude-native, isolated Anthropic project per customer. Built on a citation-first constitution: every answer either cites an approved source or flags itself as uncited. **Methodology.** - Trigger taxonomy: question patterns, frequency, repeat-asker mapping - Four-tier constitution — with the addition of a fifth state: “flag-as-uncited” - EvalOps citation-fidelity test — zero hallucinated sources allowed; release blocked on a single hallucination - Configuration loop: weekly review of “flagged-as-uncited” questions; missing docs prioritized **Evals.** Pre-deployment red-team only. Citation-fidelity benchmark numbers will be published when public-beta data is real. **Operating record.** Pre-launch. Public beta planned for Q2 2026. **Lineage.** Fidelic AI template; no specific human practitioner. ## Compatible agents - [DRYN-01 — AI Data Analyst](https://fidelic.ai/agents/dryn) — Recurring analyses, dashboard maintenance, anomaly watch - [NYRA-01 — AI Recruiter Coordinator](https://fidelic.ai/agents/nyra) — Candidate scheduling, screening, follow-up - [OMNA-01 — AI Operations Orchestrator](https://fidelic.ai/agents/omna) — Multi-agent ops coordinator for teams running 3+ agents ## Related honest questions - [What do I actually own if I cancel my AI agent tomorrow?](https://fidelic.ai/hard-questions/what-do-i-own-if-i-cancel) - [Is Fidelic just a wrapper around GPT?](https://fidelic.ai/hard-questions/wrapper-around-gpt) - [If I hire AI instead of developing my team, am I giving up on them?](https://fidelic.ai/hard-questions/developing-team) --- Canonical: https://fidelic.ai/agents/zado